#slice #unsafe #no-std #interior #immutability

no-std sealingslice

A mutable slice that can seal its initial part off from mutability, and hand out the sealed parts as immutable references

3 unstable releases

Uses old Rust 2015

0.2.0 Dec 14, 2018
0.1.1 Dec 13, 2018
0.1.0 Nov 25, 2018

#1022 in Data structures

MIT/Apache

12KB
93 lines

Sealing Slice

This crate impements a slice that can seal its initial part off from mutability, and hand out the sealed parts as immutable references.

Crate dimensions

This is a very small crate, with only about a dozen lines of effective code.

It is still provided in a dedicated crate to clearly encapsulate the guarantees of this crate from whatever uses it, and to provide some testing and stability, as opposed to ad-hoc implementations of the same concept.

All unsafety is encapsulated in the single function coalesce, which asserts that its argument slices form a contiguous area in memory and can therefore be viewed as a single slice. The SealingSlice construction around it ensures that this assertion is not violated.

When (not) to use

This crate should only be used if the standard library does not provide a safe way to the desired splitting.

An indicator that you'd actually need to use this crate is that you need to have long-lived immutable slices of your data, and later need a slice that encompasses both that earlier data and data that was still being mutated when the first one was requested.

If you merely need to split off parts of your data after you're done mutating them, and don't need to access earlier or overlapping regions thereof, slice::split_at_mut is your friend. If your splitting-off function is a method of a struct that keeps a reference to your slice, you might need to temporarily move your mutable slice out of it (and formally leave a &[] in its place) to keep the full life-times. See issue 1 for an example. In those cases, you do not need to use this crate.

Roadmap

This crate could be versioned as 1.0 soon, given there's not much interface that can change.

Variations that were considered on the API:

  • Implement the Index and IndexMut operation on the SealingSlice. Access via an immutable indexing would panic out-of-bounds when access after the seal is attempted (especially via the end-open interval for a not-completely-sealed slice), and mutable indexing would similarly err out if sealed areas are accessed.

    For users that access mutable areas at fixed offsets, this would mean less calculating their position in the current mutable view.

    Implementation would, line-of-code-wise, exceed the current code, as all variations of Range, RangeFrom, RangeToInclusive etc. would need to be covered.

  • If need for slices that seal a different part than their first bytes (eg. their tail, a single contiguous region or even an arbitrary number of areas) arises, those would make sense to be added to this crate (as the evaluation of their safety properties could be shared), but that need has not come up yet.

The former can still be implemented with this crate stabilizing; the latter is probably too different and could need to be implemented separately.

Crate status

The author's main use case for having this crate (enabling no_std use of serde-cbor has gone away since a reviewer pointed out an alternative way of implementation in issue #1. Having SealingSlice around can still be valuable, so this crate is kept alive, but -- depending on how other use cases develop -- the crate is declared as "passively maintained" for now.

No runtime deps